Displays

HS4.4

This session brings together scientists, forecasters, practitioners and stakeholders interested in exploring the use of ensemble hydro-meteorological forecast techniques in hydrological applications: e.g., flood control and warning, reservoir operation for hydropower and water supply, transportation, and agricultural management. It will address the understanding of sources of predictability and quantification and reduction of predictive uncertainty of hydrological extremes in deterministic and ensemble hydrological forecasting. Uncertainty estimation in operational forecasting systems is becoming a more common practice. However, a significant research challenge and central interest of this session is to understand the sources of predictability and development of approaches, methods and techniques to enhance predictability (e.g. accuracy, reliability etc.) and quantify and reduce predictive uncertainty in general. Ensemble data assimilation, NWP preprocessing, multi-model approaches or hydrological postprocessing can provide important ways of improving the quality (e.g. accuracy, reliability) and increasing the value (e.g. impact, usability) of deterministic and ensemble hydrological forecasts. The models involved with the methods for predictive uncertainty, data assimilation, post-processing and decision-making may include machine learning models, ANNs, catchment models, runoff routing models, groundwater models, coupled meteorological-hydrological models as well as combinations (multimodel) of these. Demonstrations of the sources of predictability and subsequent quantification and reduction in predictive uncertainty at different scales through improved representation of model process (physics, parameterization, numerical solution, data support and calibration) and error, forcing and initial state are of special interest to the session.

The session welcomes new experiments and practical applications showing successful experiences, as well as problems and failures encountered in the use of uncertain forecasts and ensemble hydro-meteorological forecasting systems. Case studies dealing with different users, temporal and spatial scales, forecast ranges, hydrological and climatic regimes are welcome.

The session is part of the HEPEX international initiative: www.hepex.org

Share:
Convener: Albrecht Weerts | Co-conveners: Shaun HarriganECSECS, Schalk Jan van Andel, Fredrik Wetterhall, Jan Verkade, Kolbjorn Engeland
Displays
| Fri, 08 May, 08:30–12:30 (CEST)

Files for download

Download all presentations (52MB)

Chat time: Friday, 8 May 2020, 08:30–10:15

D174 |
EGU2020-896
Gokcen Uysal, Rodolfo-Alvarado Montero, Dirk Schwanenberg, and Aynur Sensoy

Streamflow forecasts include uncertainties related with initial conditions, model forcings, hydrological model structure and parameters. Ensemble streamflow forecasts can capture forecast uncertainties by having spread forecast members. Integration of these forecast members into real-time operational decision models which deals with different objectives such as flood control, water supply or energy production are still rare. This study aims to use ensemble streamflows as input of the recurrent reservoir operation problem which can incorporate (i) forecast uncertainty, (ii) forecasts with a higher lead-time and (iii) a higher stability. A related technique for decision making is multi-stage stochastic optimization using scenario trees, referred to as Tree-based Model Predictive Control (TB-MPC). This approach reduces the number of ensemble members by its tree generation algorithms using all trajectories and then proper problem formulation is set by Multi-Stage Stochastic Programming. The method is relatively new in reservoir operation, especially closed-loop hindcasting experiments and its assessment is quite rare in the literature. The aim of this study is to set a TB-MPC based real-time reservoir operation with hindcasting experiments. To that end, first hourly deterministic streamflows having one single member are produced using an observed flood hydrograph. Deterministic forecasts are tested with conventional deterministic optimization setup. Secondly, hourly ensemble streamflow forecasts having a lead-time up to 48 hours are produced by a novel approach which explicitly presents dynamic uncertainty evolution. Produced ensemble members are directly provided to input to related technique. Uncertainty becomes much larger when managing small basins and small rivers. Thus, the methodology is applied to the Yuvacik dam reservoir, fed by a catchment area of 258 km2 and located in Turkey, owing to its challenging flood control and water supply operation due to downstream flow constraints. According to the results, stochastic optimization outperforms conventional counterpart by considering uncertainty in terms of flood metrics without discarding water supply purposes. The closed-loop hindcasting experiment scenarios demonstrate the robustness of the system developed against biased information. In conclusion, ensemble streamflows produced from single member can be employed to TB-MPC for better real-time management of a reservoir control system.

D175 |
EGU2020-3338
Bart van Osnabrugge, Maarten Smoorenburg, Remko Uijlenhoet, and Albrecht Weerts

There is an ongoing trend in hydrological forecasting towards both spatially distributed (gridded) models, ensemble forecasting and data assimilation techniques to improve forecasts’ initial states. While in the last years those different aspects have been investigated separately, there are only few studies where the three techniques are combined: ensemble forecasts with state updating of a gridded hydrological model. Additionally, the studies that have addressed this combination of techniques either focus on a small area, a short study period, or both. We here aim to fill this knowledge gap with a 20-year data assimilation and ensemble reforecast experiment with a high resolution gridded hydrological model (wflow_hbv, 1200x1200m) of the full Rhine basin (160 000 km2). To put the impact of state updating in an operational forecasting context, the data assimilation results were compared with AR post-processing as used by the Dutch Forecasting Centre (WMCN).

This data assimilation and reforecast experiment was conducted for the twelve main tributaries of the river Rhine. The effect on forecast skill of state updating with the Asynchronous Ensemble Kalman Filter (AEnKF) and AR error correction are compared for medium-term (15-day) forecasts over a period of 20 years (1996 to 2016). State updating improved the initial state for all subbasins and resulted in lasting skill score increase. AR also improved the forecast skill, but the forecast skill with AR did not always converge towards the uncorrected model skill, and instead can deteriorate for longer lead times. AR correction outperformed the AEnKF state updating for the first two days, after which state updating became more effective and outperformed AR. We conclude that state updating has more potential for medium-term hydrological forecasts than the operational AR procedure.

Further research is underway to investigate the importance, or added value, of long-term reforecasts as opposed to studies covering a short time span which are often more feasible and therefore more often found in literature.

D176 |
EGU2020-4233
Rahim Barzegar, Jan Adamowski, John Quilty, and Mohammad Taghi Aalami

Accurate water level (WL) forecasting is important for water resources management and planning purposes in the Great Lakes. The objectives of this research are two-fold.  The first objective is to apply machine learning (ML) (i.e., random forest (RF) and support vector regression (SVR)) and hybrid convolutional neural network(CNN)-long-short term memory (LSTM) deep learning (DL) models for multi-step (i.e., one-, two- and three-monthly step ahead) WL forecasting in the Great Lakes (Michigan and Ontario). The second objective is to integrate the boundary corrected (BC) maximal overlap discrete wavelet transform (MODWT) with SVR, RF, and CNN-LSTM models to improve the performance of the individual models. By employing a BC-wavelet decomposition method, the ‘future data’ issue (i.e., data from the future that is not available), often overlooked in the literature and a major barrier to achieving realistic forecasting performance is overcome. 

For Lakes Michigan and Ontario, 1212 monthly WL (m) records (spanning Jan 1918–Dec 2018) were used to develop the models. For the non-wavelet-based models (SVR, RF, and CNN-LSTM), candidate model inputs included the WL recorded over the previous 12 months.  For the BC-MODWT-based models (BC-MODWT-SVR, BC-MODWT-RF, and BC-MODWT-CNN-LSTM), the lagged input time series were decomposed into BC-wavelet and scaling coefficients by using different mother wavelets (Haar, Daubechies, Symlets, Fejer-Korovkin and Coiflets), filter lengths (from two up to 12) and decomposition levels (from one up to seven).  For each method (SVR, RF, and CNN-LSTM), mother wavelet, and decomposition level a model was generated.  For both wavelet- and non-wavelet-based models, the particle swarm optimization (PSO) method was used to select the most appropriate inputs to include in the proposed multi-step WL forecasting models.

The datasets were partitioned into calibration and validation subsets. After calibrating the models, various performance evaluation metrics, e.g., coefficient of determination (R2), root mean square error (RMSE), mean absolute error (MAE), root mean square percentage error (RMSPE), mean absolute percentage error (MAPE) and the Nash-Sutcliffe efficiency coefficient (NSC) were used to assess model accuracy.

Of the ML models, the SVR outperformed RF while the DL models outperformed the ML models for each forecast lead time (one-, two-, and three-step(s) ahead). Results from this case study indicate that not all wavelet families and decomposition levels perform equally and in some cases, the wavelet-based models do not improve performance over the non-wavelet-based models. However, the BC-MODWT-CNN-LSTM using suitable mother wavelets (e.g., Haar) outperforms the individual ML and BC-MODWT-ML-based models. More accurate forecasts were obtained for Lake Michigan although the performance in both Great Lakes was accurate. The outcomes of this research indicate that the BC-MODWT-CNN-LSTM model is a promising tool for generating accurate WL forecasts.

D177 |
EGU2020-5036
Trine Jahr Hegdahl, Kolbjørn Engeland, Ingelin Steinsland, and Andrew Singleton

In this work the performance of different pre- and postprocessing methods and schemes for ensemble forecasts were compared for a flood warning system.  The ECMWF ensemble forecasts of temperature (T) and precipitation (P) were used to force the operational hydrological HBV model, and we estimated 2 years (2014 and 2015) of daily retrospect streamflow forecasts for 119 Norwegian catchments. Two approaches were used to preprocess the temperature and precipitation forecasts: 1) the preprocessing provided by the operational weather forecasting service, that includes a quantile mapping method for temperature and a zero-adjusted gamma distribution for precipitation, applied to the gridded forecasts, 2)  Bayesian model averaging (BMA) applied to the catchment average values of temperature and precipitation. For the postprocessing of catchment streamflow forecasts, BMA was used. Streamflow forecasts were generated for fourteen schemes with different combinations of the raw, pre- and postprocessing approaches for the two-year period for lead-time 1-9 days.

The forecasts were evaluated for two datasets: i) all streamflow and ii) flood events. The median flood represents the lowest flood warning level in Norway, and all streamflow observations above median flood are included in the flood event evaluation dataset. We used the continuous ranked probability score (CRPS) to evaluate the pre- and postprocessing schemes. Evaluation based on all streamflow data showed that postprocessing improved the forecasts only up to a lead-time of 2 days, while preprocessing T and P using BMA improved the forecasts for 50% - 90% of the catchments beyond 2 days lead-time. However, with respect to flood events, no clear pattern was found, although the preprocessing of P and T gave better CRPS to marginally more catchments compared to the other schemes.

In an operational forecasting system, warnings are issued when forecasts exceed defined thresholds, and confidence in warnings depends on the hit and false alarm ratio. By analyzing the hit ratio adjusted for false alarms, we found that many of the forecasts seemed to perform equally well. Further, we found that there were large differences in the ability to issue correct warning levels between spring and autumn floods. There was almost no ability to predict autumn floods beyond 2 days, whereas the spring floods had predictability up to 9 days for many events and catchments.

The results underline differences in the predictability of floods depending on season and the flood generating processes, i.e. snowmelt affected spring floods versus rain induced autumn floods. The results moreover indicate that the ensemble forecasts are less good at predicting correct autumn precipitation, and more emphasis could be put on finding a better method to optimize autumn flood predictions. To summarize we find that the flood forecasts will benefit from pre-/postprocessing, the optimal processing approaches do, however, depend on region, catchment and season.

D178 |
EGU2020-6047
Pallav Kumar Shrestha, Christof Lorenz, Husain Najafi, Stephan Thober, Oldrich Rakovec, and Luis Samaniego

Semi-arid regions are characterized by low annual precipitation that exhibit large seasonal fluctuations. While semi-arid regions cover 3.6% of the globe, 13% of world’s documented reservoirs (GRanD database) are within 100 km of semi-arid regions to fulfill water demand year-round. Reservoirs are known to increase evaporation and significantly change hydrologic regime downstream. Accurate representation of reservoirs and scale independent modeling is indispensable for reliable hydrologic forecasting systems in semi-arid regions. To address this, the mesoscale hydrological model (mHM, git.ufz.de/mhm) is augmented with a new lake/reservoir module (multiscale lake module, mLM). The objective is to measure the performance of a scalable seasonal forecasting model chain with and without reservoirs.

The experimental setup constitutes the SaWaM (http://grow-sawam.org/) project study regions encompassing seven semi-arid basins and 15 reservoirs of high significance across three continents (Sao Francisco, Jaguaribe, Piranhas in Brazil, Blue Nile, Atbara in Sudan, Karun in Iran, Chira-Catamayo in Ecuador).The calibration of mHM parameters and its initial conditions for forecsating are obtained using the spatially disaggregated ERA5 (ERA-SD, ≈ 10 km, starting 1981) climate reanalysis data. The calibrated model is forced with an ensemble of 25 realisations of ECMWF-SEAS5 seasonal hindcasts which are bias corrected and spatially disaggregated (BCSD, ≈10 km) using ERA-SD. The 2010–2016 hindcasting experiment generates hydrological forecasts with lead time of upto six months. The performance of the model chain BCSD-mHM-mLM and BCSD-mHM are evaluated using the Brier Skill Score.

Preliminary results show that incorporating reservoirs in the model improves the performance of mHM (average NSE improvement ≈ +0.1 for the period 1990–2010) and the overarching forecasting model chain. Sub-grid level lake delineation and in-/outflow calculations of mLM result in scalable reservoir states and fluxes and thus overall scalable basin hydrology. Seamless forecasts for soil moisture, streamflow, reservoir inflow and reservoir water level are achieved across scales (≈10 km to ≈1 km) showing skills to up to two months lead time. This study is the first step towards an operational hydrological seasonal forecasting system which has potential to significantly improve water management, specially in semi-arid regions.

D179 |
EGU2020-8166
Jean Odry, Marie-Amélie Boucher, Simon Lachance Cloutier, Richard Turcotte, and Pierre-Yves Saint-Louis

In snow-prone regions, snowmelt is one of the main drivers of runoff. For operational flood forecasting and mitigation, the spatial distribution of snow water equivalent (SWE) in near real time is necessary. In this context, in situ observations of SWE provide a valuable information. Nonetheless, the high spatial variability of snowpack characteristics makes it necessary to implement some kind of snow modelling to get a spatially continuous estimation. Data assimilation is thus a useful approach to combine information from both observation and modeling in near real-time.

For example, at the provincial government of Quebec (eastern Canada), the HYDROTEL Snowpack Model is applied on a daily basis over a 0.1 degree resolution mesh covering the whole province. The modelled SWE is corrected in real time by in situ manual snow survey which are assimilated using a spatial particles filter (Cantet et al., 2019). This assimilation method improves the reliability of SWE estimation at ungauged sites.

The availability of manual snow surveys is however limited both in space and time. These measurements are conducted on a bi-weekly basis in a limited number of sites. In order to further improve the temporal and spatial observation coverage, alternative sources of data should be considered.

In this research, it is hypothesized that data gathered by SR50 sonic sensors can be assimilated in the spatial particle filter to improve the SWE estimation. These automatic sensors provide hourly measurements of snow depth and have been deployed in Quebec since 2005. Beforehand, probabilistic SWE estimations were derived from the SR50 snow depth measurements using an ensemble of artificial neural networks (Odry et al. 2019). Considering the nature of the data and the conversion process, the uncertainty associated with this dataset is supposed larger than for the manual snow surveys. The objective of the research is to evaluate the potential interest of adding this lower-quality information in the assimilation framework.

The addition of frequent but uncertain data in the spatial particle filter required some adjustments in term of assimilation frequency and particle resampling. A reordering of the particles was implemented to maintain the spatial coherence between the different particles. With these changes, the consideration of both manual snow surveys and SR50 data in the spatial particle filter reached performances that are comparable to the initial particle filter that combines only the model and manual snow survey for estimating SWE in ungauged sites. However, the addition of SR50 data in the particle filter allows for continuous information in time, between manual snow surveys.

 

References:

Cantet, P., Boucher, M.-A., Lachance-Coutier, S., Turcotte, R., Fortin, V. (2019). Using a particle filter to estimate the spatial distribution of the snowpack water equivalent. J. Hydrometeorol, 20.

Odry, J., Boucher, M.-A., Cantet,P., Lachance-Cloutier, S., Turcotte, R., St-Louis, P.-Y. (2019). Using artificial neural networks to estimate snow water equivalent from snow depth. Canadian water ressources journal (under review)

D180 |
EGU2020-13223
Steven Weijs and Hossein Foroozand

Probabilistic forecasts are essential for good decision making, because they communicate the forecaster's best attempt at representation of both information available and the remaining uncertainty of a variable of interest. The amount of information provided, which can be measured in bits using information theory, would then be a natural measure of success for the forecast in a verification exercise. On the other hand, it may seem rational to tune the forecasting system to provide maximum value to users. Somewhat counter-intuitively, there are arguments against tuning for maximum value. When the design of the forecasting system also includes the choice of the sources of information, monitoring network optimization becomes part of the problem to solve.  
In this presentation, we give a brief overview of the different roles information theory can have in these different aspects of probabilistic forecasting. These roles range from analysis of predictability, model selection, forecast verification, monitoring network design, and data assimilation by ensemble weighting. Using the same theoretical framework for all these aspects has the advantage that some connections can be made that may eventually lead to a more unified perspective on forecasting. 

D181 |
EGU2020-12069
Seong Jin Noh, James McCreight, Moha El Gharamti, Tim Hoar, Arezoo Rafieeinasab, and Benjamin Johnson

The Data Assimilation Research Testbed (DART) has been coupled with the community WRF-Hydro modeling system with the intent of providing efficient and flexible support for assimilating a wide range of streamflow and soil moisture observations and delivering an ensemble of model states useful for quantifying streamflow uncertainties. The coupled framework, named Hydro-DART, is used to study and assess the flooding consequences of Hurricane Florence over the Carolinas during August-September 2018 period.
Several extensions to earlier versions of Hydro-DART have been explored. These include: (1) a multi-configuration ensemble in which different ensemble members are run with different physical parameters (e.g., Manning's roughness and channel geometry) in order to create additional ensemble variability, (2) a variable transform, anamorphosis, which is introduced such that bounded quantities (e.g., streamflow) are transformed to a Gaussian space prior to the Kalman update as a way to avoid non-physical state updates, (3) a spatially-correlated noise, which is introduced to represent uncertainty of input forcings (e.g., overland and subsurface fluxes) in a physically meaningful way, and (4) an along-the-stream localization, which considers precipitation correlation length scale, rather than physical proximity. Hourly streamflow gauge data, from the flood-affected area, is used to test the impact of these extensions on the overall prediction accuracy. Analyses and hindcasts are compared to those based on the nudging assimilation currently employed in the National Water  Model (NWM) operations. Standard streamflow forecast metrics are also supplemented by a wavelet-based event timing error metric.

D182 |
EGU2020-11428
Kian Abbasnezhadi and Alain N. Rousseau

The applicability of the Canadian Precipitation Analysis products known as the Regional Deterministic Precipitation Analysis (CaPA-RDPA) for hydrological modelling in boreal watersheds in Canada, which are constrained with shortage of precipitation information, has been the subject of a number of recent studies. The northern and mid-cordilleran alpine, sub-alpine, and boreal watersheds in Yukon, Canada, are prime examples of such Nordic regions where any hydrological modelling application is greatly scrambled due to lack of accurate precipitation information. In the course of the past few years, proper advancements were tailored to resolve these challenges and a forecasting system was designed at the operational level for short- to medium-range flow and inflow forecasting in major watersheds of interest to Yukon Energy. This forecasting system merges the precipitation products from the North American Ensemble forecasting System (NAEFS) and recorded flows or reconstructed reservoir inflows into the HYDROTEL distributed hydrological model, using the Ensemble Kalman Filtering (EnKF) data assimilation technique. In order to alleviate the adverse effects of scarce precipitation information, the forecasting system also enjoys a snow data assimilation routine in which simulated snowpack water content is updated through a distributed snow correction scheme. Together, both data assimilation schemes offer the system with a framework to accurately estimate flow magnitudes. This robust system not only mitigates the adverse effects of meteorological data constrains in Yukon, but also offers an opportunity to investigate the hydrological footprint of CaPA-RDPA products in Yukon, which is exactly the motivation behind this presentation. However, our overall goal is much more comprehensive as we are trying to elucidate whether assimilating snow monitoring information in a distributed hydrological model could meet the flow estimation accuracy in sparsely gauged basins to the same extent that would be achieved through either (i) the application of precipitation analysis products, or (ii) expanding the meteorological network. A proper answer to this question would provide us with valuable information with respect to the robustness of the snow data assimilation routine in HYDROTEL and the intrinsic added-value of using CaPA-RDPA products in sparsely gauged basins of Yukon.

D183 |
EGU2020-19132
Maarten Smoorenburg, Klaudia Horváth, Tjerk Vreeken, Ruben Sinnige, Stefan Nieuwenhuis, Rodolfo Alvarado-Montero, Teresa Piovesan, and Peter Gijsbers

Decision making in operational water management practice is particularly challenging during extreme events. Dealing with extreme events would typically benefit from longer anticipation times, yet forecast uncertainty is often large for extreme events, and grows with lead time. Classical Model Predictive Control (MPC) only considers one deterministic forecast (no uncertainty), making control in anticipation of extreme events highly susceptible to forecast biases. MPC methods that can represent forecast uncertainty through ensemble techniques have been developed, but are rarely used in practice due to the mathematical complexity and computational burden.
We set out to test whether newly developed mathematically rigorous implementations of two ensemble based MPC methods could contest this status quo; one method that takes into account that new information comes available in the future and can be acted upon (i.e., the control tree approach of Raso et al., 2014), and one that does not.  We conducted a set of closed-loop experiments with synthetic forecasts of inflow and storm surge, and compared the control results of the ensemble based MPC methods to control with deterministic MPC. We did this for varying degrees of forecast uncertainty and bias. The experiments were conducted for the Volkerak-Zoommeer lake in the Netherlands, a simple example of a water system where water levels should be maintained within a narrow bandwidth by operating drainage works that only allow outflow to sea at low tide. An event with simultaneous high inflows and storm surge at sea can here only be mitigated by timely creation of retention capacity through lowering of the lake level.
The control of such an extreme event was mimicked with each MPC method by computing a single optimal control strategy every 12h (but looking 5 days ahead), and simulating the resulting lake level to obtain starting conditions for the next control time in 12h. All models and methods were implemented within the Python-based open source MPC software framework RTC-Tools 2, allowing fast and robust convex optimization of water systems. Since the control of the outlet requires boolean decision variables to account for the flow direction —typically boosting computation times—, advanced linearization techniques were needed to keep computation times short enough for operational practice.
The experiments showed that the ensemble based MPC methods can more robustly control the lake level than deterministic MPC, which with even mildly underestimating forecasts resulted in worse mitigation of the event. The ensemble method without control tree, known to be more conservative, could provide better control, but, for large forecasts uncertainties, did so by lowering the lake level too much. This illustrates that deciding upon which ensemble method to use requires choices about how conservative the controller should be.
The experiments also demonstrate that it is feasible to use ensemble forecasts in combination with ensemble based MPC methods in operational water management practice. This opens doors to including uncertainty information in the operational decision making process in objective ways. More details about the optimization and ensemble techniques are presented in session HS3.3 by Horváth et al., 2020.

D184 |
EGU2020-18694
Gaia Piazzi, Guillaume Thirel, and Charles Perrin

Skillful streamflow forecasts provide a key support to several water-related applications. Ensemble forecasting systems are gaining a widespread interest, since they allow accounting for different sources of uncertainty. Because of the critical impact of the initial conditions (ICs) on the forecast accuracy, it is essential to improve their estimates via data assimilation (DA). This study aims at assessing the sensitivity of the DA-based estimation of forecast ICs to several sources of uncertainty and to the update of different model states and parameters of a conceptual rainfall-runoff model. The performance of two sequential ensemble-based techniques are compared, namely Ensemble Kalman filter and Particle filter, in terms of both efficiency and temporal persistence of the updating effect through the assimilation of observed discharges at the forecast time. Several experiments specifically address the impact of the meteorological, model state and parameter uncertainties over 232 catchments in France. Results show that the benefit of the DA-based estimation of ICs for forecasting is the largest when focusing on the level of the model routing store, which is the internal state the most correlated to streamflow. While the EnKF-based forecasts outperform the PF-based ones when accounting for the meteorological uncertainty, the representation of the model state uncertainty allows greatly improving the accuracy of the PF-based predictions, with a longer-lasting updating effect (up to 10 days). Conversely, the forecasting skill is undermined when accounting for the parameter uncertainty, due to the change in the hydrological responsiveness through the update of both the production and routing store levels. A further effort is focused on assessing the impact of the spatial resolution of the hydrological model on the predictive accuracy of DA-based streamflow forecasts.

D185 |
EGU2020-3597
Wouter Greuell and Ronald Hutjes

This contribution deals with the skill of a physical model-based system built to produce probabilistic seasonal hydrological forecasts, applied here to South America and earlier to Europe (see  Greuell et al., hess-23-371-2019). The system basically consists of the Variable Infiltration Capacity (VIC) hydrological model forced with output from ECMWF’s Seasonal Forecasting System 5 (SEAS5). We analyse skill in runoff and discharge hindcasts both with real observations and with so-called pseudo-observations, i.e. with discharge data generated with VIC forced with historical meteorological observations (1981-2015). At the continental scale discrimination skill in runoff shows characteristics that are similar to Europe. Especially, even at the longest lead time (7 months) significant skill remains in 20-30% of both continents. However, in the first lead month there is less significant skill in South America, due to absence of skill in its very dry and very wet regions, than in Europe, where similar extremes do not exist. To explain the skill in runoff, we performed two suites of specific hydrological hindcasts akin to Ensemble Streamflow Predictions (ESP), which each isolate a different source of skill (meteorological forcing and initial conditions). We find that in South America the contribution to skill by forcing is larger than in Europe, which can be ascribed to differences in the skill in the precipitation forcing. Even at a lead time of 7 months, the precipitation hindcasts have significant skill in 15-30% of South America while in Europe skill is almost confined to the first lead month. Discharge hindcasts for grid cells with a sufficient amount of observations were post-processed with ensemble model output statistics (EMOS). This procedure successfully increased reliability but resulted in a small decrease of discrimination skill. Nevertheless, for the location of the Itaipu dam, used to produce 18% of Brazil’s electricity, discrimination skill is highly significant for the post-processed discharge, e.g. at all lead times in the last two months of the year.

D186 |
EGU2020-19564
Artur Safin, Damien Bouffard, James Runnalls, Fotis Georgatos, Eric Bouillet, Firat Ozdemir, Fernando Perez Cruz, and Jonas Šukys

Lakes form an integral component of ecosystems and our communities. Aside from being a source of drinking water, lakes provide additional benefits such as recreation, heat management and fishing. At the same time, human activity can significantly disrupt the natural state of the aquatic ecology. In the past, limited understanding of the hydrological and biochemical processes in aquatic systems has led to significant eutrophication in certain cases. To mitigate further risk, monitoring programs have been implemented. Recently new instrumentation, such as in situ observation platforms, remote sensing and computational resources enable comprehensive monitoring of the temporal evolution of the environment’s spatial heterogeneity.

A major focus of the DATALAKES project is to use the multiple sources of observational measurements for data assimilation and forecasting purposes. The aim is to infer the entire state of the lake as accurately as possible using high-resolution three-dimensional hydrodynamic models. Uncertainty quantification using Bayesian inference and modern Markov Chain Monte Carlo methods is implemented using the SPUX package, with the stochasticity provided by an ensemble of weather forecasts. To obtain predictions in a reasonable time, we parallelize both the particle filtering and the hydrodynamic model on the CSCS cluster. The data assimilation component will combine multiple in-situ sources with remote sensing measurements of lake water surface temperature and incorporate the respective uncertainties in measurement into the error model. To enable the use of multi-level variance reduction schemes, we perform calibration of essential hydrodynamic model parameters for a hierarchy of discretisations. The results show that the framework is capable of inferring the state of lake Geneva from observational measurements.

D187 |
EGU2020-17754
Andrea Ficchi, Hannah Cloke, Ervin Zsoter, Christel Prudhomme, and Liz Stephens

Severe flooding in southern Africa is caused by a variety of meteorological hazards including intense tropical cyclones and depressions, mesoscale convective complexes and persistent lows which bring extreme rainfall and flood events with different characteristics. Little is known about the relative predictability of flooding associated to these different drivers, especially in operational forecasting systems. Understanding the limits of predictability for the different drivers of flooding is important to provide evidence of forecast capabilities to end-users and decision-makers and build trust in the use of the forecasting systems.

Here we explore the skill of probabilistic flood forecasts from the operational Copernicus Emergency Management Service Global Flood Awareness System (GloFAS v2) over southern Africa. GloFAS provides real-time hydrological forecasts up to 30 days ahead by coupling ensemble weather forecasts from the European Centre for Medium-Range Weather Forecasts (ECMWF) with hydrological modelling. The GloFAS flood forecasts are openly available and can support humanitarians and other international organisations to trigger action before a devastating flood occurs.

Using hydrological records of past flood events over the last 20 years, the GloFAS forecast skill is assessed by analysing the probability of detection of the events over different lead-times from 1 to 30 days, as well as the consistency and accuracy of predictions of event-based characteristics such as the flood timing and duration. We stratify the analysis by the multi hazard drivers of flooding with a focus on the distinction between tropical cyclones and other types of meteorological events. We suggest that such a stratified analysis of forecast skill can help modellers better understand the sources of predictability in flood forecasts and can support humanitarians to define specific trigger levels for forecast-based action for different types of flood events.

D188 |
EGU2020-1369
Teng Zhang, Zhongjing Wang, and Zixiong Zhang

Runoff forecast with high precision is important for the efficient utilization of water resources and regional sustainable development, especially in the arid area. The monthly runoff of Changmabao (CMB) station has an upwards trend and an abrupt point in 1998. The impact factor analysis shows that it is highly correlated with the current precipitation and temperature in the wet season while the previous runoff and previous global land temperature in the dry season. Three models including the time-series decomposition model, the model based on teleconnection coupled with the support vector machine, and the model based on teleconnection coupled with the artificial neural network are used to predict the runoff of CMB station. An indicator β is constructed with the correlation coefficient (R) and mean relative deviation (rBias) to evaluate the model performance more conveniently and intuitively. The results suggest that the model based on teleconnection coupled with the support vector machine preforms best. This forecasting method could be applied to the management and dispatch of water resources in arid areas.

D189 |
EGU2020-2828
Yuchen Liu, Jia Liu, Chuanzhe Li, Fuliang Yu, Wei Wang, and Qingtai Qiu

    WRF-Hydro is not only a stand-alone hydrological modeling architecture but also a coupling component for integrating hydrological models with atmospheric models. Sensitivity tests are carried out in this study for the most important parameters influencing the streamflow generation of the WRF/WRF-Hydro coupled system by targeting at the semi-humid and semi-arid catchments in Northern China. The main objective of the study is the parameters controlling the total water volume and the shape of the hydrograph are refined on the basis of sensitivity tests and their effects on the generation of the streamflow are addressed with the intent to apply the modeling system for streamflow forecasting. Two major aspects are considered in the calibration process for testing the sensitivity of the WRF-Hydro model parameters. On the one hand, it is to consider the parameters controlling the total water volume, which include the runoff infiltration parameter (REFKDT), and the surface retention depth (RETDEPRT) controlled by a scaling parameter named RETDEPRTFAC. One the other hand, it is to look at the parameters controlling the shape of the hydrograph, which include the channel Manning roughness parameter (MannN), and the overland flow roughness parameter (OVROUGHRT) controlled by the scaling parameter OVROUGHRTFAC. Through the sensitivity tests of the parameters affecting the runoff, it is found that REFKDT and MannN are the most sensitive parameter especially with unsaturated soil conditions. The findings of this study is to explore the variation laws of the key parameters in semi-humid and semi-arid areas, and to provide a reference for calibration and application of the WRF/WRF-Hydro coupled system.

D190 |
EGU2020-2879
Rana Muhammad Adnan Ikram, Zhongmin Liang, Ozgur Kisi, Muhammad Adnan, Binquan Li, and Kuppusamy Sathishkumar

River runoff prediction plays a very vital role in water resources planning, hydropower designing and agricultural water management. In the current study, the prediction capability of three machine learning models, least square support vector regression (LSSVR), fuzzy genetic (FG) and M5 model tree (M5Tree), in modeling daily and monthly runoffs of Hunza River catchment (HRC) using own and nearby Gilgit climatic station data is examined. The prediction performances of three machine learning models are compared using three statistical indexes, namely, root mean square error (RMSE), mean absolute error (MAE) and coefficient of determination (R2). Firstly, four previous time lagged values of runoff, rainfall and atmospheric temperature are used as inputs on basis of correlation analysis to validate and test the accuracy of three machine learning models. After analyzing the performance of various input combinations, optimal one is selected for each variable and then these optimal inputs are employed together to see the forecasting performance. In the first part of study, monthly runoff of HRC are predicted using inputs consisting of local previous monthly runoff values and monthly meteorological values of Gilgit station. The test results show that LSSVR provides more accurate prediction results than the other two machine learning models. In the second part, daily runoffs of HRC are predicted using own previous daily runoff and Gilgit station’s climatic values. In the test results, a better accuracy is obtained from LSSVR models in relative to the FG and M5Tree models. In the last part of study, daily runoffs of HRC are predicted using own runoff and climatic data of HRC. In the results, it is found that local climatic data slightly improved the all model’s prediction accuracy in comparison of other scenario which also uses nearby station’s climatic data. The LSSVR models again are found to be better than the FGA and M5Tree models. LSSVM generally performs superior to the FGA and M5Tree in forecasting daily stream flow of Hunza River using local stream flow and climatic inputs. Based on the results of study, LSSVR model is recommended for monthly and daily runoff prediction of HRC with or without local climatic data.

D191 |
EGU2020-8842
Vsevolod Moreydo, Boris Gartsman, Valentina Khan, and Vladimir Tischenko

We present the post-processing technique for the operational ensemble forecasting system (EFS) currently applied to the Cheboksary reservoir on the Volga River in Russia. The operational forecasting system is built around the ECOMAG semi-distributed hydrological model and has shown to produce reliable forecasts of spring snowmelt water inflow into the reservoir on lead-times up to four months ahead (Gelfan et al., 2018). We propose the improvement of the mean reservoir monthly inflow forecast skill by constructing cumulative distributions (CDF) of the observed streamflow conditioned on the predicted streamflow from the EFS and observed mean monthly air temperature and precipitation. We overcome the limitation of short time-series of the observed variables by multivariate modelling procedure allowing for the time-series extension. The extended time-series are then classified into 64 categories each containing the unique combination of the predictors by their quartile values, and the observed monthly inflow CDFs are constructed. The improved operational forecast CDF is consequently picked from the obtained 64 CDF classes by defining the appropriate CDF class from the combination of the raw ensemble forecast and any weather prediction available. The proposed technique was assessed by using the SL-AV weather model (Khan, 2011; Tolstykh, 2017) monthly temperature and precipitation hindcasts for the evaluation period of 1982 – 2010. The forecasts were benchmarked against climate and observed (perfect) weather forecast and have shown improvement in terms of reliability and resolution.

The research is supported by the Russian Science Foundation, project no. 17-77-30006.

References:

Gelfan, A., Moreydo, V., Motovilov, Y., & Solomatine, D. P. (2018). Long-term ensemble forecast of snowmelt inflow into the Cheboksary Reservoir under two different weather scenarios. Hydrology and Earth System Sciences, 22(4). https://doi.org/10.5194/hess-22-2073-2018

Tolstykh, M., Shashkin, V., Fadeev, R., and Goyman, G.: Vorticity-divergence semi-Lagrangian global atmospheric model SL-AV20: dynamical core, Geosci. Model Dev., 10, 1961–1983, https://doi.org/10.5194/gmd-10-1961-2017, 2017

Khan V.M., Kryzhov V.N., Vilfand R.M., Tishchenko V.A., Bundel A.Y. Multimodel approach to seasonal prediction. Russian Meteorology and Hydrology. 2011. Т. 36. № 1. С. 11-17.

Chat time: Friday, 8 May 2020, 10:45–12:30

D192 |
EGU2020-9492
Adele Young, Biswa Bhattacharya, and Chris Zevenbergen

Pluvial flooding is on the rise as more cities are challenged by a changing climate and local drivers: increased urbanisation and inadequate sewer system capacity. Flood forecasting and early warning systems have been proposed as a “low regret” measure to reduce flood risk and increase preparedness through forecast-based actions.  However there are multiple sources of uncertainty from meteorological forecast, model parameters and structure and inadequate calibration.  In data-scarce cities, there are additional challenges to produce high-quality rainfall forecast and well-calibrated flood forecast (timing, water levels, extent and impact). As a result, there is a cascading effect on the ability to make and provide good reliable decisions given the uncertainty in the forecast or inaccuracy in the input data.

Ensemble prediction systems (EPS) have been proposed as a means to quantify uncertainty in forecast and compared to deterministic forecast, facilitate a probabilistic framework in decision making. Probabilistic information has been applied to cost loss ratio approaches and Bayesian decision under uncertainty. However, to what extent inherent spatiotemporal inaccuracies of meteorological inputs influence this posterior probability and the resultant decision has not been considered in data scare regions. In this regard, this research focuses on providing understanding on how the influence of the varying degrees of input data, particularly forecast rainfall spatial and temporal distributions will ultimately affect the ability to make an optimal decision; i.e. the recommended decision given the information available at the time of the forecast.

Using a study area in the Alexandria city, Egypt, this research proposes a framework for decision making under uncertainty in an urban data-scarce city using a Weather Research Forecast (WRF) model to simulate downscaled rainfall ensemble forecast and remotely sensed rainfall products to supplement data gaps. Adopting a probabilistic approach, uncertainty in the flood forecast predictions will be represented from an urban rainfall-runoff model driven by ensemble precipitation forecast. The objective of this research is not to make forecast more accurate but rather to highlight the interdependences of the flood forecast and decision-making chain in order to address what decision can be made given the quality of forecast.

Keywords: Pluvial flood forecasting, Ensemble forecast, Decision making, Data-scarce Alexandria, Egypt

D193 |
EGU2020-10845
Eduardo Muñoz-Castro, Pablo A. Mendoza, and Ximena Vargas

In catchments with a highly variable flow regime, an accurate and reliable hydrological forecasting framework is critical to support water resources management. However, due to model structural deficiencies and changing climatic conditions, the parameter estimates during the calibration period are expected to vary with hydrological conditions. This work aims to test the added value of incorporating potential non-stationarities in hydrologic model parameters on seasonal streamflow forecasts in high-mountain environments, using the ensemble streamflow prediction (ESP) methodology. To this end, we apply the GR4J rainfall-runoff model coupled with the snow accumulation and ablation CemaNeige module in six basins located in Central Chile (30-36° S). We explore the effects of four parameter selection strategies on the quality of seasonal streamflow forecasts produced with the ESP method: (i) a single set of parameters for the entire hindcast period (our benchmark), (ii) using parameters calibrated with a ‘leave-one-year-out’ approach, (iii) using parameter sets based on expected hydroclimatic conditions, and (iv) dual data assimilation to improve the initial condition and parameters before the forecast initialization. Results show that parameters related to production store capacity in GR4J model, and degree-day melt coefficient and weighting coefficient for snow pack thermal state in the CemaNeige module have a high inter-annual variability, with variations of 50% with respect to the benchmark scenario.

D194 |
EGU2020-11457
Zhenwu Wang, Rolf Hut, and Nick van de Giesen

The eWaterCycle program provides a collaborative environment for hydrological modelers, developed by the Netherlands eScience Center together with the Delft University of Technology. It aims to build a community of scientists in hydrology who use different programming languages for their specific models. Python is the lingua franca of the eWaterCycle platform and requires no modification to a particular model, making the platform user-friendly and flexible. Therefore, it can readily be applied in other geoscientific models. Currently, the Python data assimilation package includes ensemble-type methods, particle filters, and their variants, which are all sequential data assimilation algorithms. The implementation of techniques related to localization and inflation methods is included in this package. Localization and inflation are effective ways to avoid the collapse of a filter, which happens commonly in high dimensional models. The package gives access to all tunable parameters by configuration files quickly. To evaluate the performance of data assimilation comprehensively, a series of metrics is provided. In addition, the package offers a set of visualization tools to explore the results of data assimilation and the improvement of models.

D195 |
EGU2020-11890
Hanoi Medina and Di Tian

Soil moisture forecasting is important for informing agricultural and environmental management. However, due to the strong interactions between climate, soils and vegetation, even small errors in the weather-related forcing commonly have remarkable impacts on the soil moisture, making soil moisture forecasting especially challenging. Therefore, it is necessary to develop a probabilistic forecasting strategy that accounts for the uncertainty of inputs such as rainfall and evapotranspiration. Here we develop a hybrid dynamic-statistical framework that combines statistical downscaled forecasts of precipitation and reference evapotranspiration from numerical weather predictions (NWP) with a probabilistic water balance model to produce probabilistic forecasts for daily soil moisture at site scale. Forecasts are initialized using in situ measurements over represented locations from the National Soil Moisture Network. We found that the skill of the soil moisture forecasts more relies on the skill of the precipitation forecasts than reference evapotranspiration forecasts. The forecasts are invariably highly skillful over the first 2-3 days, while the skill rapidly decreases over the following days. The soil moisture forecasts based on the statistically post-processed NWP forecasts show higher skill than persistence-based forecasts, climatology forecasts, or forecasts directly retrieved from NWP.

D196 |
EGU2020-12604
Naoki Koyama and Tadashi Yamada

The aim of this paper is to verify the accuracy of the real-time flood prediction model, using the time-series analysis. Forecast information of water level is important information that encourages residents to evacuate. Generally, flood forecasting is conducted by using runoff analysis. However, in developing countries, there are not enough hydrological data in a basin. Therefore, this study assumes where poor hydrologic data basin and evaluates it through reproducibility and prediction by using time series analysis which statistical model with the water level data and rainfall data. The model is applied to the one catchment of the upper Tone River basin, one of the first grade river in Japan. This method is possible to reproduce hydrograph, if the observation stations exist several points in the basin. And using the estimated parameters from past flood events, we can apply this method to predict the water level until the flood concentration time which the reference point and observation station. And until this time, the peak water level can be predicted with the accuracy of several 10cm. Prediction can be performed using only water level data, but by adding rainfall data, prediction can be performed for a longer time.

D197 |
EGU2020-12734
Huihui Dai

The formation of runoff is extremely complicated, and it is not good enough to forecast the future runoff only by using the previous runoff or meteorological data. In order to improve the forecast precision of the medium and long-term runoff forecast model, a set of forecast factor group is selected from meteorological factors, such as rainfall, temperature, air pressure and the circulation factors released by the National Meteorological Center  using the method of mutual information and principal component analysis respectively. Results of the forecast in the Qujiang Catchment suggest the climatic factor-based BP neural network hydrological forecasting model has a better forecasting effect using the mutual information method than using the principal component analysis method.

D198 |
EGU2020-11756
Di Tian, Parisa Asadi, Hanoi Medina, Brenda Ortiz, and Isaya Kesikka

A key challenge for climate-smart water management is timely and reliably forecasting potential evapotranspiration (ETc) and irrigation water requirement (IWR) at field level with high spatial and temporal resolution. In this study, we develop a framework for forecasting ETc and IWR using multi-model numerical weather predictions and harmonized Landsat Sentinel-2 remote sensing product. Multiple numerical weather predictions from The International Grand Global Ensemble (TIGGE) are used as input into the Food and Agriculture Organization (FAO) Penman-Monteith equation to produce reference evapotranspiration (ETo) forecasts. The non-homogeneous Gaussian regression method is used to post-process the ETo forecasts. ETo forecasts are evaluated against meteorological observations and compare with the forecasts from the National Weather Service Digital Forest Database over the contiguous United States. Crop parameters (leaf area index and surface albedo) and crop coefficients are derived from visible and near-infrared images from Harmonized Landsat Sentinel-2 product. The satellite derivations are evaluated against ground crop measurements from agricultural fields in Alabama and California. Potential crop evapotranspiration (ETc) forecasts are estimated using two approaches: 1) crop coefficient-based approach, and 2) crop parameter-based approach with the post-processed ETo forecasts. The ETc and irrigation water requirement (IWR) calculated using the FAO-56 method with observed weather data and field collected crop data are used as observational reference. ETc and IWR forecasts are evaluated against observational references using different metrics. In general, statistical post-processing using non-homogeneous Gaussian regression greatly improved ETo forecasting performance. The crop parameter-based approach showed better performance compared to the crop coefficient approach, contingent upon the choice of TIGGE predictions. The study demonstrates the capability of Harmonized Landsat Sentinel-2 and TIGGE for forecasting ETc and IWR and has implications for informing site-specific climate smart water management.  

D199 |
EGU2020-13232
Jan Verkade, Fredrik Wetterhall, Maria-Helena Ramos, Andy Wood, Quan Wang, Ilias Pechlivanidis, and Marie-Amélie Boucher

Since 2004, HEPEX (Hydrologic Ensemble Prediction Experiment) has built a community of researchers and practitioners around the world. After 15 years, its mission continues to be very relevant: to establish a more integrated view of hydrologic forecasting. In this view, data assimilation, hydro-meteorological modelling chains, user behavioural-decision models, pre- and post-processing techniques, expert knowledge, participatory co-evolution of knowledge and user needs, communication and visualisation tools, training material, games and decision support systems are connected to enhance operational services, early warning systems and water management applications. Great progress has been made over the years in terms of using ensemble hydro-meteorological forecasting, but there are still institutional, scientific and operational challenges that the community faces. Here, we present the full range of HEPEX activities, such as workshops, conference sessions, testbeds, learning material and our long-running portal (www.hepex.org). We show how HEPEX can continue to be a relevant network in the coming decades. A large part of that answer lies in the fact that our members use the platform to continuously share their research, make announcements, report on workshops, projects and meetings, and hear about related research and operational challenges. It is also a forum for early career scientists to become increasingly involved in hydrologic forecasting science and applications.

D200 |
EGU2020-16683
Schalk Jan van Andel

Most continuous verification metrics for hydrometeorological forecasts are based on equal interval forecasts and observations (e.g. daily, 6-hourly, etc.). For some purposes of verification, however, it might be more beneficial to have variable time intervals that take into account the duration of events, e.g. rainfall or flood event (or discharge exceeding a flood warning threshold). Such verification, however, is challenged by defining the length of the non-event intervals for scoring correct rejections and false alarms, needed for continuous verification.  The work presented here suggests how to approach this challenge and presents verification results of a continuous forecast verification method that take into account variable duration of events.

D201 |
EGU2020-15220
Jens Grundmann, Achim Six, and Andy Philipp

Reliable warnings and forecasts of extreme precipitation and the resulting floods are an important prerequisite for disaster response. Especially for small catchments, warning and forecasting systems are challenging due to the short response time of the catchments and the uncertainties of the meteorological forecast products. Thus, ensemble forecasts of precipitation are an option to portray these inherent uncertainties. By this contribution, we present our operational processing scheme for ensemble hydrological forecasting. We use the COSMO-D2-EPS product of the German Weather Service, which provides an ensemble of 20 members each three hours, for lead times up to 27 hours. Each member is evaluated regarding specific extreme precipitation thresholds for predefined hydrological warning regions. If these thresholds are exceeded in a specific region, rainfall-runoff models for the associated catchments are started to propagate the meteorological uncertainty into the resulting runoff, followed by statistical post processing and visualization. In addition, a communication and training concept based on a series of workshops with the locally responsible civil protection forces to deal with the uncertainties in the forecast is associated. Results are presented by a re-analysis of the flood in the upper Weiße Elster catchment in May 2018 in the Vogtland region of Saxony. Rainfall amounts larger than 140mm in 6 hours led to the highest flood warning levels in the region. Analysis show that such extreme amounts of precipitation are only predicted by one member of the COSMO-D2-EPS ensemble forecast. The deterministic COSMO-D2 model run does not show this, which underlines the benefit and the potential of the ensemble predictions, but also the need for a suitable communication of the uncertainties.

D202 |
EGU2020-13573
Ignacio Martin Santos, Mathew Herrnegger, Hubert Holzmann, Kristina Fröhlich, and Jennifer Ostermüller

In the last years, the demand of reliable seasonal streamflow forecasts has increased with the aim of incorporating them into decision support systems for e.g. river navigation, power plant operation  or drought risk management. Recently, the concept of “climate services” has gained stronger attention in Europe, thereby incorporating useful information derived from climate predictions and projections that support adaptation, mitigation and disaster risk management. In the frame of one of these climate services currently in development, Clim2Power project, a seasonal forecast system for discharge in the Upper Danube upstream Vienna has been established.

Seasonal forecasts are generated using a dynamical approach running a hydrological model (COSERO) with forecasted climate input provided by DWD (Germany's National Meterological Service). The climate forecasts are based on a large ensemble of predictions, available up to 6 months. After the application of a statistical downscaling method, the climate forecasts have a spatial resolution of 6km. The predictability is related to two main contributions: meteorological forcings (i.e. temperature and precipitation predictability) and initial basin states at the time the forecast is issued.

The Upper Danube basin with a catchment area of approx. 100.000 km2 is characterized by complex topography dominated by the Alps, elevations range from about 150 m to slightly under 4000 m. Therefore, the skill of the seasonal forecast is highly influenced by the resolution of the meteorological data, and likewise by the hydrological processes that take place, especially, regarding melting processes. Downscaled hindcasts over the last 20 years, generated with the identical setup as the seasonal forecasts, are used in this contribution to assess the skill of the seasonal forecasts. In addition, some post-processing corrections, based on historical observations, are used to adjust the bias of the forecasts. Nevertheless, remaining non-systematic error patterns do not allow complete bias correction. Apart from the biases, also the correlation patterns show a limited skill. We conclude that the seasonal discharge forecasting is still not sufficient to incorporate the results into water resources decision support systems within the studied Alpine basins.

D203 |
EGU2020-18472
Gwyneth Matthews, Hannah Cloke, Sarah Dance, and Christel Prudhomme

Floods are the most common and disastrous natural hazards and due to climate change and socio-economic growth they are becoming more dangerous. Early warning systems are one of the best ways to decrease the effect of floods by increasing preparedness. The European Flood Awareness System (EFAS), part of the European Commission's Copernicus Emergency Management Service, provides medium-range ensemble flood forecasts for the whole of Europe but only calibrates its forecasts locally at river gauge stations where sufficiently long and reliable observations are available. These corrections do not consider the natural relationships that occur between points up and downstream. In this PhD project, data assimilation techniques will be used, in post-processing, to combine the available gauge observations with the forecasts. Using a weighting matrix, the influence of the observations will be extended along the river channel network, taking account of the ensemble and observation uncertainty. EFAS uses meteorological forcings from four numerical weather prediction (NWP) systems, so a multi-model approach will need to be developed.  This requires new data assimilation theory and hydrological process knowledge to ensure consistent updates. Delocalising calibrations will improve the accuracy of forecasts at unobserved locations allowing end-users to make more informed decisions to mitigate flood damage.

D204 |
EGU2020-19883
Alper Onen, Mehdi H. Afshar, and Burak Bulut

This study investigates the performance of short term daily hydrological forecasts utilizing Global Forecast System (GFS) and Hydrologiska Byrans Vattenbalansavdelning (HBV) model over a major sub-region (~25000 km2) located in Euphrates River Basin, Turkey. The test basin, over which the forecast algorithm is implemented, is home to five (four operational and another almost-complete) dams with 3 three more planned; all eight located within last 300 km reach of 730 km-long Murat River. The algorithm is aimed to operate through a user-friendly and reliable commercially available forecast interface for decision makers working on fields such as energy production scheme optimization and flood mitigation. In the development of this forecast strategy, the main basin was divided into multiple subbasins that are bordered with corresponding facilities and each subbasin is independently modelled & calibrated by HBV model and meteorological records using available stream gauge and weather station data collected in the region. The Global Forecast System (GFS) that provides 16-day meteorological forecast is then applied to model as the input for hydrological predictions. By implementing and clustering daily operation records of inflow and outflow data received directly from related regional SCADA system, volumetric hydrograph corrections are applied on the model output as a final corrective filter to maximize the temporal performance over the 16-day forecast period. As the system has been in use for nearly 20 months (as of January 2020), our results have shown that a calibrated data cluster performance nearing 0.98 has been reached in correlation and 0.90 in Nash-Sutcliffe index:  cluster-independent average weekly inspected forecast performance of almost 0.87 in correlation and 0.85 in Nash-Sutcliffe index has been obtained.

D205 |
EGU2020-20645
Fredrik Wetterhall, Umberto Modigliani, Milan Dacic, and Sari Lappi

The project “South-East European Multi-Hazard Early Warning Advisory System” (SEE-MHEWS-A) is a collaborative effort to strengthen the existing early warning capacity in the region. The project was initiated in 2014 by the World Meteorological Organization (WMO), and has been supported by the U.S. Agency for International Development (USAID) and the World Bank. The project will test a prototype of a flood early warning system using local information and multiple models to better characterize the flood risk in selected catchments. The aims of the project are: (1) is to strengthen regional co-operation by leveraging national, regional and global capacities to develop improved hydrometeorological forecasts, advisories and warnings to save lives and limit economic losses, (2) strengthen national multi-hazard early warning systems by making tools and data available to the participating countries and other beneficiaries, (3) implement impact-based forecasts and risk-based warnings utilizing non-deterministic hydrometeorological modeling to support governments, disaster management authorities, humanitarian agencies, non-governmental organizations, and other stakeholders in their decision-making and actions, and (4) to harmonise forecasts and warnings in trans-boundary areas. During 18 month the project will setup a full-hydrometeorological forecasting system, including observational data storage and sharing, limited area modelling of the meteorological forcing data and hydrological forecasting.

D206 |
EGU2020-22659
Emixi Valdez, Francois Anctil, and Maria-Helena Ramos

Skillful hydrological forecasts are essential for decision-making in many areas such as preparedness against natural disasters, water resources management, and hydropower operations. Despite the great technological advances, obtaining skillful predictions from a forecasting system, under a range of conditions and geographic locations, remain a difficult task. It is still unclear why some systems perform better than others at different temporal and spatial scales. Much work has been devoted to investigate the quality of forecasts and the relative contributions of meteorological forcing, catchment’s initial conditions, and hydrological model structure in a streamflow forecasting system. These sources of uncertainty are rarely considered fully and simultaneously in operational systems, and there are still gaps in understanding their relationship with the dominant processes and mechanisms that operate in a given river basin. In this study, we use a multi-model hydrological ensemble prediction system (H-EPS) named HOOPLA (HydrOlOgical Prediction Laboratory), which allows to account separately for these three main sources of uncertainty in hydrological ensemble forecasting. Through the use of EnKF data assimilation, of 20 lumped hydrological models, and of the 50-member ECMWF medium-range weather forecasts, we explore the relationship between the skill of ensemble predictions and the many descriptors (e.g. catchment surface, climatology, morphology, flow threshold and hydrological regime) that influence hydrological predictability. We analyze streamflow forecasts at 50 stations spread across Quebec, France and Colombia, over the period from 2011 to 2015 and for lead times up to 9 days. The forecast performance is assessed using common metrics for forecast quality verification, such as CRPS, Brier skill score, and reliability diagrams. Skill scores are computed using a probabilistic climatology benchmark, which was generated with the hydrological models forced by resampled historical meteorological data. Our results contribute to relevant literature on the topic and bring additional insight into the role of each descriptor in the skill of a hydrometeorological ensemble forecasting chain, serving as a possible guide for potential users to identify the circumstances or conditions in which it is more efficient to implement a given system.

 

D207 |
EGU2020-13684
Marko Kallio, Joseph H.A. Guillaume, Alexander J. Horton, and Timo A. Räsänen

Global climate and hydrological modelling have shown that human influence on the hydrosphere has been growing and is projected to continue increasing. Global models can inform us of the regional trends and events occurring in the stream network, however, operational water management and research often require tailored and detailed modelling to support decision making. Decisions on which kind of hydrological model (lumped, distributed) and at what scale can, however, impact on the usability of the model outputs for use cases which were not anticipated during the model set-up.

Here we conduct two experiments with an objective to determine whether an ensemble of a downscaled Global Hydrological Models (GHM) can be used 1) to improve the performance, and 2) to spatially disaggregate the output of a catchment scale model to its sub-basins. We use two existing distributed models set up for research purposes in the Sekong, Sesan, and Srepok Rivers (a major tributary of the Mekong), and in the Grijalva-Usumacinta catchments in Mexico. In the first experiment, we downscale off-the-shelf runoff products from the Inter-Sectoral Impact Model Intercomparison Project (ISIMIP) using a recently developed areal interpolation method, route the downscaled runoff, and apply model averaging on an ensemble consisting of the downscaled GHM timeseries and the output of the distributed model at the observation stations. In the second experiment, we downscale and route runoff from the GHMs down the river network, as in the first experiment. During the routing step we record the sub-basin of origin and the timestep of runoff as it reaches an observation station. This record is then used to reconstruct a distributed estimate of discharge (back-traced from the existing model output) in all river reaches. We validate the reconstructed distributed estimates by comparing their spatial distribution to the outputs of the original distributed hydrological models, and against streamflow records.

Our initial experiments show that the downscaled estimates from GHMs have potential to increase the performance of the model outputs. We also show that the reconstruction of hydrographs in sub-basins of the modelled area is possible, however, the uncertainties related to the method are large and the estimates are sensitive to the routing solution used in the back-tracing, and to the performance of the ensemble of GHMs.

The methodology has potential for improving the usability of GHMs in local contexts. Owing to the promptly available GHM outputs, the method allows for swift exploration of hydrological questions before a proper modelling experiment is set up. Using GHMs as supplementary ensemble members can also aid in locations where calibration of the models is difficult due to scarce or ill-fitting data, or when the original choice of model fails to capture some aspects of the hydrograph.

D208 |
EGU2020-20187
Bastian Klein, Ilias Pechlivanidis, Louise Arnal, Louise Crochemore, Dennis Meissner, and Barbara Frielingsdorf

Many sectors, such as hydropower, agriculture, water supply and waterway transport, need information about the possible evolution of meteorological and hydrological conditions in the next weeks and months to optimize their decision processes on a long term. With increasing availability of meteorological seasonal forecasts, hydrological seasonal forecasting systems have been developed all over the world in the last years. Many of them are running in operational mode. On European scale the European Flood Awareness System EFAS and SMHI are operationally providing seasonal streamflow forecasts. In the context of the EU-Horizon2020 project IMPREX additionally a national scale forecasting system for German waterways operated by BfG was available for the analysis of seasonal forecasts from multiple hydrological models.

Statistical post processing tools could be used to estimate the predictive uncertainty of the forecasted variable from deterministic / ensemble forecasts of a single / multi-model forecasting system. Raw forecasts shouldn’t be used directly by users without statistical post-processing because of various biases. To assess the added potential benefit of the application of a hydrological multi-model ensemble, the forecasting systems from EFAS, SMHI and BfG were forced by re-forecasts of the ECMWF’s Seasonal Forecast System 4 and the resulting seasonal streamflow forecasts have been verified for 24 gauges across Central Europe. Additionally two statistical forecasting methods - Ensemble Model Output Statistics EMOS and Bayesian Model Averaging BMA - have been applied to post-process the forecasts.

Overall, seasonal flow forecast skill is limited in Central Europe before and after post-processing with a current predictability of 1-2 months. The results of the multi-model analysis indicate that post-processing of raw forecasts is necessary when observations are used as reference. Post-processing improves forecast skill significantly for all gauges, lead times and seasons. The multi-model combination of all models showed the highest skill compared to the skill of the raw forecasts and the skill of the post-processed results of the individual models, i.e. the application of several hydrological models for the same region improves skill, due to the different model strengths.